Zellen aus Karzinomen sind erwiesenermaßen weicher als Epithelzellen ihres Ursprungsgewebes. Es wurde vermutet, dass dieses Weicherwerden Zellen dabei hilft, aus dem Primärtumor auszubrechen und Metastasen zu bilden, was allerdings erst von wenigen Belegen bestärkt wird. Weiterhin wird die Entwicklung von Karzinomen allgemein als von einer epithelial-mesenchymalen Transition (EMT) angetrieben angesehen, ein Prozess, der die Umformung von Epithelgeweben steuert und stark in das Zytoskelett eingreift. Ich habe daher die Hypothese aufgestellt, dass EMT Karzinomzellen weicher macht und somit aggressive und invasive Tumore erzeugt. In der vorgelegten Arbeit gehe ich dem Nachweis dieser Hypothese nach. Ich habe den Einfluss der EMT auf Zellweichheit in vitro untersucht, allerdings kein gerichtetes Weicherwerden mit Fortschritt der EMT feststellen können. Mit vitalen Einzelzellen, die ich aus Operationsresektaten isoliert habe, verglich ich die mechanischen Eigenschaften von invasiven und nicht-invasiven Tumoren ex vivo und konnte eine klare Korrelation von Aggressivität mit Zellweichheit in vier verschiedenen Arten von Karzinomen aufzeigen. Membrangebundenes E-cadherin, das mir als Marker für den Fortschritt der EMT diente, war jedoch weder mit der Aggressivität der Karzinome noch mit der Weichheit derer Zellen korreliert. Ich benutzte maschinelles Lernen (ML), um Krebs-zellen in silico auf Basis ihrer mechanischer Eigenschaften zu klassifizieren, stieß aber auf klare Grenzen. In dieser Arbeit habe ich zum ersten Mal ex vivo gezeigt, dass das Weicherwerden von Krebszellen ein kontinuierlicher Prozess in Karzinomen ist, und dass erhöhte Aggressivität mit erhöhter Zellweichheit einhergeht. Ich habe außerdem EMT, die lange Zeit als entscheidend für Zellinvasion galt, als mögliche Ursache für dieses Weicherwerden ausgeschlossen. Zusammengenommen mit meinen Resultaten der ML Klassifikation deutet dies darauf hin, dass eine erhöhte Heterogenität von mechanischen Eigenschaften von Krebszellen, ausgelöst von allgemeiner Deregulation, die Invasion von Karzinomen antreibt.:1 Introduction 1 2 Background 11 2.1 The cytoskeleton of eukaryotic cells 12 2.2 The actin-E-cadherin-complex 17 2.2.1 E-cadherin 17 2.2.2 The Wnt/β-catenin pathway 18 2.2.3 Actin-E-cadherin dynamics 19 2.3 The epithelial to mesenchymal transition (EMT) 21 2.3.1 Epithelial and mesenchymal cells 21 2.3.2 Classical EMT 22 2.3.3 EMT in carcinoma development 23 2.4 Carcinoma development 25 2.4.1 Growth and spread 25 2.4.2 Tumor grading and staging 26 2.4.3 Carcinoma development outside of EMT 29 2.5 Cell mechanics in migration and invasion 31 3 Materials & methods 37 3.1 The Optical Stretcher as a main measurement device for cellular softness and E-cadherin level 38 3.1.1 Deformation by radiation pressure 39 3.1.2 Viability in an OS 43 3.1.3 Data acquisition and evaluation 46 3.2 Kelvin Voigt (KV) modeling 50 3.3 Machine learning 53 3.3.1 Interpreting and evaluating classications 54 3.3.2 Data preparation 58 3.3.3 Support vector machines (SVM) 58 3.3.4 Random forest (RF) 64 3.3.5 Permutation importance 67 3.4 Statistical analysis 68 3.4.1 Two one-sided tests (TOST) as a statistical test for equivalence 69 3.5 In vitro model systems for eukaryotic cells, their culture, and preparation 71 3.5.1 Cell lines 71 3.5.2 Cell culture 73 3.5.3 Fluorescent labeling of E-cadherin 73 3.6 Isolation of cancer cells from primary samples 75 3.6.1 Isolation of cancer cells from blood samples 75 3.6.2 Isolation of cancer cells from surgical resections 77 4 Results & discussion 79 4.1 In vitro growth factor induced EMT 81 4.1.1 EGF induced EMT is not correlated to cell softening in MCF 10A epithelial cells 82 4.1.2 TGFβ1 induced EMT is not correlated to cell softening in MCF 10A epithelial cells 87 4.1.3 Summary 91 4.2 Ex vivo vital tumor cells from liquid biopsies and surgical resections 94 4.2.1 Database analysis reveals that there is no systematic change of EMT related markers over the course of carcinoma progression 96 4.2.2 Vital single cells isolated from liquid biopsies of breast cancer patients can be distinguished from healthy cells of their natural surrounding 99 4.2.3 Cell softening is correlated to aggressiveness in tumor cells isolated from surgical resections 110 4.2.4 EMT progression is connected to neither cell softening nor aggressiveness in tumor cells isolated from surgical resections 120 4.2.5 Summary 123 4.3 In silico Machine learning as means to assess the predictive power of cell mechanics 127 4.3.1 Parameters from OS measurements 128 4.3.2 In vitro discrimination of cell types in a breast cancer cell line panel 129 4.3.3 Ex vivo discrimination of breast cancer cells and PBMC isolated from liquid biopsies 136 4.3.4 Summary 143 5 Conclusion & outlook 147 A Additional data and information 161 A.1 Optimization of support vector machines (SVM) and random forest (RF) machine learning approaches 161 A.1.1 Optimization of the training set size in SVM and RF machine learning approaches 161 A.1.2 Optimization of the SVM machine learning algorithm 161 A.1.3 Optimization of the RF machine learning algorithm 163 A.2 List of features for machine learning based classication 164 A.2.1 List of features used for classication of my in vitro cell line panel 164 A.2.2 List of features for classication of circulating tumor cells isolated from the blood of patients with mamma carcinoma 166 A.3 Activity parameter A of cells isolated from the blood samples of breast cancer patients 170 B Materials and reagents 171 B.1 Cell culture media 171 B.1.1 Medium for MCF 10A cells 171 B.1.2 Medium for MDA-MB-436 and MDA-MB-231 cells 171 B.1.3 Medium for NIH/3T3 cells 172 B.2 Ringer lactate buer for tissue transport and storage 172 B.3 MACS buffer 172 C Protocols 173 C.1 In vitro culture of cell lines 173 C.1.1 Passage of cell lines cultured in vitro 173 C.1.2 Cryogenic storage and thawing of cell lines 174 C.2 Immunouorescent labeling of E-cadherin 174 C.3 Growth factor treatment of MCF 10A epithelial cells 175 C.3.1 Treatment with increasing concentrations of epidermal growth factor (EGF) 175 C.3.2 Treatment with constant concentration of epidermal growth factor (EGF) 176 C.3.3 Treatment with transforming growth factor β1 (TGFβ1) 177 C.4 Isolation of vital cells from patient samples 178 C.4.1 Negative depletion of specic populations from cell suspensions by magnetic bead sorting 178 C.4.2 Isolation of vital circulating tumor cells (CTC) from the blood of patients with mamma carcinoma 179 C.4.3 Isolation of healthy peripheral blood mononuclear cells (PBMC) from the blood of patients and donors 179 C.4.4 Isolation of vital cancer cells from tumor samples of surgical resections of various carcinomas 180 C.5 Immunohistochemical staining of paranized tissue slices of tumor tissue 82 Bibliography 186 / Carcinoma cells have been shown to be softer than cells from their tissue of origin, healthy epithelia. This softening effect has been predicted to drive tumor cell migration and ergo metastases, but only circumstantial evidence exists for this. Carcinoma development is also generally viewed as driven by an epithelial to mesenchymal transition (EMT), a process that governs epithelial restructuring and heavily interferes with the cytoskeleton. I therefore hypothesized that EMT drives cell softening in carcinomas, which in turn leads to aggressive and invasive tumors. In the presented work, I pursue the verification of this hypothesis. I investigated the influence of EMT on cell softening in vitro, yet found no directed development of cell body softness with EMT progression. With vital single cancer cells that I isolated from surgical resections, I explored the mechanics of invasive, and non-invasive tumors ex vivo and saw a clear correlation of tumor aggressiveness with cell softness in four different types of carcinomas. There was however no correlation between E-cadherin in the cell membrane of isolated cancer cells, which I used as a marker for EMT progression, and the aggressiveness of the respective carcinomas or the softness of their cells. I employed machine learning (ML) to classify cancer cells based on their mechanical properties in silico, but found clear limits to that approach. In this work, I have shown for the very first time ex vivo how cell softening is an ongoing process during carcinoma development and increased aggressiveness is linked to increased softness. I also excluded EMT, which has long been deemed a driver of cell invasion, as a possible origin for cell softening. Together with results from ML classification, this points to increased heterogeneity in mechanical properties of cancer cells by deregulation as a main contributor to carcinoma invasion.:1 Introduction 1 2 Background 11 2.1 The cytoskeleton of eukaryotic cells 12 2.2 The actin-E-cadherin-complex 17 2.2.1 E-cadherin 17 2.2.2 The Wnt/β-catenin pathway 18 2.2.3 Actin-E-cadherin dynamics 19 2.3 The epithelial to mesenchymal transition (EMT) 21 2.3.1 Epithelial and mesenchymal cells 21 2.3.2 Classical EMT 22 2.3.3 EMT in carcinoma development 23 2.4 Carcinoma development 25 2.4.1 Growth and spread 25 2.4.2 Tumor grading and staging 26 2.4.3 Carcinoma development outside of EMT 29 2.5 Cell mechanics in migration and invasion 31 3 Materials & methods 37 3.1 The Optical Stretcher as a main measurement device for cellular softness and E-cadherin level 38 3.1.1 Deformation by radiation pressure 39 3.1.2 Viability in an OS 43 3.1.3 Data acquisition and evaluation 46 3.2 Kelvin Voigt (KV) modeling 50 3.3 Machine learning 53 3.3.1 Interpreting and evaluating classications 54 3.3.2 Data preparation 58 3.3.3 Support vector machines (SVM) 58 3.3.4 Random forest (RF) 64 3.3.5 Permutation importance 67 3.4 Statistical analysis 68 3.4.1 Two one-sided tests (TOST) as a statistical test for equivalence 69 3.5 In vitro model systems for eukaryotic cells, their culture, and preparation 71 3.5.1 Cell lines 71 3.5.2 Cell culture 73 3.5.3 Fluorescent labeling of E-cadherin 73 3.6 Isolation of cancer cells from primary samples 75 3.6.1 Isolation of cancer cells from blood samples 75 3.6.2 Isolation of cancer cells from surgical resections 77 4 Results & discussion 79 4.1 In vitro growth factor induced EMT 81 4.1.1 EGF induced EMT is not correlated to cell softening in MCF 10A epithelial cells 82 4.1.2 TGFβ1 induced EMT is not correlated to cell softening in MCF 10A epithelial cells 87 4.1.3 Summary 91 4.2 Ex vivo vital tumor cells from liquid biopsies and surgical resections 94 4.2.1 Database analysis reveals that there is no systematic change of EMT related markers over the course of carcinoma progression 96 4.2.2 Vital single cells isolated from liquid biopsies of breast cancer patients can be distinguished from healthy cells of their natural surrounding 99 4.2.3 Cell softening is correlated to aggressiveness in tumor cells isolated from surgical resections 110 4.2.4 EMT progression is connected to neither cell softening nor aggressiveness in tumor cells isolated from surgical resections 120 4.2.5 Summary 123 4.3 In silico Machine learning as means to assess the predictive power of cell mechanics 127 4.3.1 Parameters from OS measurements 128 4.3.2 In vitro discrimination of cell types in a breast cancer cell line panel 129 4.3.3 Ex vivo discrimination of breast cancer cells and PBMC isolated from liquid biopsies 136 4.3.4 Summary 143 5 Conclusion & outlook 147 A Additional data and information 161 A.1 Optimization of support vector machines (SVM) and random forest (RF) machine learning approaches 161 A.1.1 Optimization of the training set size in SVM and RF machine learning approaches 161 A.1.2 Optimization of the SVM machine learning algorithm 161 A.1.3 Optimization of the RF machine learning algorithm 163 A.2 List of features for machine learning based classication 164 A.2.1 List of features used for classication of my in vitro cell line panel 164 A.2.2 List of features for classication of circulating tumor cells isolated from the blood of patients with mamma carcinoma 166 A.3 Activity parameter A of cells isolated from the blood samples of breast cancer patients 170 B Materials and reagents 171 B.1 Cell culture media 171 B.1.1 Medium for MCF 10A cells 171 B.1.2 Medium for MDA-MB-436 and MDA-MB-231 cells 171 B.1.3 Medium for NIH/3T3 cells 172 B.2 Ringer lactate buer for tissue transport and storage 172 B.3 MACS buffer 172 C Protocols 173 C.1 In vitro culture of cell lines 173 C.1.1 Passage of cell lines cultured in vitro 173 C.1.2 Cryogenic storage and thawing of cell lines 174 C.2 Immunouorescent labeling of E-cadherin 174 C.3 Growth factor treatment of MCF 10A epithelial cells 175 C.3.1 Treatment with increasing concentrations of epidermal growth factor (EGF) 175 C.3.2 Treatment with constant concentration of epidermal growth factor (EGF) 176 C.3.3 Treatment with transforming growth factor β1 (TGFβ1) 177 C.4 Isolation of vital cells from patient samples 178 C.4.1 Negative depletion of specic populations from cell suspensions by magnetic bead sorting 178 C.4.2 Isolation of vital circulating tumor cells (CTC) from the blood of patients with mamma carcinoma 179 C.4.3 Isolation of healthy peripheral blood mononuclear cells (PBMC) from the blood of patients and donors 179 C.4.4 Isolation of vital cancer cells from tumor samples of surgical resections of various carcinomas 180 C.5 Immunohistochemical staining of paranized tissue slices of tumor tissue 82 Bibliography 186

Einer der Gründe für das Scheitern derzeitiger Behandlungsmethoden beim Brustkrebs ist die Resistenz gegenüber der angewandten Chemotherapie. Eine große Rolle bei der Entstehung der Multiplen Medikamentenresistenz spielt das MDR1-Gen und sein Genprodukt, das P-Glykoprotein. Das Y-Box-Protein YB-1 reguliert die Expression des MDR1-Gens; eine Überexpression und nukleäre Lokalisation von YB-1 geht im Brustkrebs mit einer gesteigerten P-Glykoprotein Expression einher. In dieser Arbeit wurden Gewebeproben von 83 Brustkrebspatientinnen auf eine YB-1 Überexpression im Tumor und im peritumoralen Epithel untersucht. YB-1 wurde mittels der immunhistochemischen APAAP-Methode an Formalin-fixierten, in Paraffin eingebetteten Brustkrebsgewebeproben nachgewiesen. Die klinische Relevanz der YB-1 Expression wurde untersucht, indem sie mit dem klinischen Verlauf in einem mittleren Beobachtungszeitraum von 61 Monaten und etablierten biologischen Tumorfaktoren wie Lymphknotenstatus, histologisches Grading, Tumorgröße, Hormonrezeptorstatus, uPA und PAI-1 verglichen wurde. In der Kohorte der Patientinnen mit einer postoperativen adjuvanten Chemotherapie zeigte sich eine 5-Jahres-Rezidivrate von 68 % bei einer hohen YB-1 Expression im Tumor und eine Rückfallrate von 39 % bei einer niedrigen YB-1 Expression. Unter Beachtung auch der YB-1 Expression im peritumoralen Epithel konnte ein noch größerer Unterschied hinsichtlich der 5-Jahres-Rezidivrate festgestellt werden. Diese betrug bei Patientinnen mit einer hohen YB-1 Expression 66 %, während bei Patientinnen mit einer niedrigen YB-1 Expression im Nachbeobachtungszeitraum kein Rezidiv festgestellt wurde. Bei der Gegenüberstellung der 5-Jahres-Rezidivraten in der Kohorte der Patientinnen ohne Zytostatikatherapie zeigte sich eine Rückfallrate von 30 % bei einer hohen YB-1 Expression und eine Rückfallrate von 0 % bei einer niedrigen YB-1 Expression. Eine hohe YB-1 Expression war demnach in beiden Kohorten mit einer schlechteren klinischen Prognose assoziiert. Das Ergebnis in der Gruppe der Patientinnen ohne postoperative Chemotherapie zeigt, dass YB-1 mit der Tumoraggressivität beim Brustkrebs korreliert. Eine Korrelation zwischen der YB-1 Expression und den etablierten prognostischen Faktoren Lymphknotenstatus, Tumorgröße und histologisches Grading konnte nicht festgestellt werden. Es wurde jedoch eine signifikante negative Korrelation zwischen der YB-1 Expression und dem Hormonrezeptorstatus und eine positive Korrelation zwischen YB-1 und den Faktoren uPA und PAI-1 gefunden. In dieser Arbeit wurde gezeigt, dass YB-1 eine klinische Relevanz besitzt mit Hinblick sowohl auf eine prognostische als auch eine prädiktive Bedeutung bei der Identifikation von Hoch-Risiko-Patientinnen im Brustkrebs in Ab- und Anwesenheit einer postoperativen Chemotherapie. / Intrinsic or acquired resistance to chemotherapy is one of the reasons for failure of current treatment regimens in breast cancer patients. P-glycoprotein and its gene mdr1 plays a major role in the development of a multi-drug resistant tumor phenotype. The Y-box protein YB-1 regulates the expression of mdr1. In human breast cancer, overexpression and nuclear localization is associated with upregulation of P-glycoprotein. In this study, tissues of 83 breast cancer patients have been analyzed with regard to YB-1 overexpression in tumor tissue and in surrounding benign breast epithelial cells. YB-1 has been detached by the immunohistochemical APAAP-method using formalin-fixed, paraffin-embedded breast cancer tissues. Clinical relevance of YB-1 expression was analyzed by comparing it with clinical outcome after a median follow-up of 61 months and with tumor biological factors lymph-node status, tumor size, histological grading, hormone-receptor status and the factors uPA and PAI-1. In patients who received postoperative chemotherapy, the 5-year-relapse rate was 68% in patients with high YB-1 expression in tumor cells and 39% in patients with low expression. With regard to YB-1 expression in surrounding benign breast epithelial cells, the 5-year-relapse rate was 66% in patients with high YB-1 expression whereas in patients with low expression no relapse has been observed so far. YB-1 thus indicates clinical drug resistance in breast cancer. In patients who received no chemotherapy, the 5-year-relapse rate was 30% in patients with high YB-1 expression whereas in patients with low YB-1 expression no relapse occurred. YB-1 thus correlates with breast cancer aggressiveness. In both groups high YB-1 expression was associated with poor clinical outcome. A correlation between YB-1 and tumor biological factors lymph-node status, tumor size and histological grading has not been found. But a significant negative correlation has been observed between YB-1 and hormone-receptor status and a positive correlation between YB-1 and uPA and PAI-1. This dissertation could show the clinical relevance of YB-1 with regard to a prognostic and predictive significance by identifying a high-risk group of breast cancer patients both in presence and absence of postoperative chemotherapy.